🤖 AI Summary
This study addresses the challenge of lung adenocarcinoma grading, which relies on accurate identification of the predominant growth pattern but typically demands costly, slide-level fine-grained annotations. To mitigate this requirement, the authors propose an attention-based multiple instance learning (MIL) framework that operates solely with whole-slide image-level labels. The approach leverages a pretrained foundation model in computational pathology—such as Prov-GigaPath—as a patch encoder and employs an attention mechanism to aggregate global features for growth pattern prediction. This work represents the first integration of a pathology foundation model with attention-based MIL, substantially reducing dependence on pixel-level annotations while enhancing prediction robustness. Experimental results demonstrate that fine-tuned Prov-GigaPath within the ABMIL framework achieves state-of-the-art performance (Cohen’s κ = 0.699), significantly outperforming conventional aggregation baselines.
📝 Abstract
Lung adenocarcinoma (LUAD) grading depends on accurately identifying growth patterns, which are indicators of prognosis and can influence treatment decisions. Common deep learning approaches to determine the predominant pattern rely on patch-level classification or segmentation, requiring extensive annotations. This study proposes an attention-based multiple instance learning (ABMIL) framework to predict the predominant LUAD growth pattern at the whole slide level to reduce annotation burden. Our approach integrates pretrained pathology foundation models as patch encoders, used either frozen or fine-tuned on annotated patches, to extract discriminative features that are aggregated through attention mechanisms. Experiments show that fine-tuned encoders improve performance, with Prov-GigaPath achieving the highest agreement (\k{appa} = 0.699) under ABMIL. Compared to simple patch-aggregation baselines, ABMIL yields more robust predictions by leveraging slide-level supervision and spatial attention. Future work will extend this framework to estimate the full distribution of growth patterns and validate performance on external cohorts.